xref: /freebsd/cddl/usr.sbin/zfsd/zfsd.cc (revision 3fc36ee018bb836bd1796067cf4ef8683f166ebc)
1 /*-
2  * Copyright (c) 2011, 2012, 2013, 2014, 2015, 2016  Spectra Logic Corporation
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions, and the following disclaimer,
10  *    without modification.
11  * 2. Redistributions in binary form must reproduce at minimum a disclaimer
12  *    substantially similar to the "NO WARRANTY" disclaimer below
13  *    ("Disclaimer") and any redistribution must be conditioned upon
14  *    including a substantially similar Disclaimer requirement for further
15  *    binary redistribution.
16  *
17  * NO WARRANTY
18  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
19  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
20  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
21  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
22  * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
26  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
27  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
28  * POSSIBILITY OF SUCH DAMAGES.
29  *
30  * Authors: Justin T. Gibbs     (Spectra Logic Corporation)
31  */
32 
33 /**
34  * \file zfsd.cc
35  *
36  * The ZFS daemon consumes kernel devdctl(4) event data via devd(8)'s
37  * unix domain socket in order to react to system changes that impact
38  * the function of ZFS storage pools.  The goal of this daemon is to
39  * provide similar functionality to the Solaris ZFS Diagnostic Engine
40  * (zfs-diagnosis), the Solaris ZFS fault handler (zfs-retire), and
41  * the Solaris ZFS vdev insertion agent (zfs-mod sysevent handler).
42  */
43 
44 #include <sys/cdefs.h>
45 #include <sys/param.h>
46 #include <sys/fs/zfs.h>
47 
48 #include <err.h>
49 #include <libgeom.h>
50 #include <libutil.h>
51 #include <poll.h>
52 #include <syslog.h>
53 
54 #include <libzfs.h>
55 
56 #include <list>
57 #include <map>
58 #include <string>
59 
60 #include <devdctl/guid.h>
61 #include <devdctl/event.h>
62 #include <devdctl/event_factory.h>
63 #include <devdctl/exception.h>
64 #include <devdctl/consumer.h>
65 
66 #include "callout.h"
67 #include "vdev_iterator.h"
68 #include "zfsd_event.h"
69 #include "case_file.h"
70 #include "vdev.h"
71 #include "vdev_iterator.h"
72 #include "zfsd.h"
73 #include "zfsd_exception.h"
74 #include "zpool_list.h"
75 
76 __FBSDID("$FreeBSD$");
77 
78 /*================================== Macros ==================================*/
79 #define NUM_ELEMENTS(x) (sizeof(x) / sizeof(*x))
80 
81 /*============================ Namespace Control =============================*/
82 using DevdCtl::Event;
83 using DevdCtl::EventFactory;
84 using DevdCtl::EventList;
85 
86 /*================================ Global Data ===============================*/
87 int              g_debug = 0;
88 libzfs_handle_t *g_zfsHandle;
89 
90 /*--------------------------------- ZfsDaemon --------------------------------*/
91 //- ZfsDaemon Static Private Data ----------------------------------------------
92 ZfsDaemon	    *ZfsDaemon::s_theZfsDaemon;
93 bool		     ZfsDaemon::s_logCaseFiles;
94 bool		     ZfsDaemon::s_terminateEventLoop;
95 char		     ZfsDaemon::s_pidFilePath[] = "/var/run/zfsd.pid";
96 pidfh		    *ZfsDaemon::s_pidFH;
97 int		     ZfsDaemon::s_signalPipeFD[2];
98 bool		     ZfsDaemon::s_systemRescanRequested(false);
99 EventFactory::Record ZfsDaemon::s_registryEntries[] =
100 {
101 	{ Event::NOTIFY, "DEVFS", &DevfsEvent::Builder },
102 	{ Event::NOTIFY, "GEOM",  &GeomEvent::Builder },
103 	{ Event::NOTIFY, "ZFS",   &ZfsEvent::Builder }
104 };
105 
106 //- ZfsDaemon Static Public Methods --------------------------------------------
107 ZfsDaemon &
108 ZfsDaemon::Get()
109 {
110 	return (*s_theZfsDaemon);
111 }
112 
113 void
114 ZfsDaemon::WakeEventLoop()
115 {
116 	write(s_signalPipeFD[1], "+", 1);
117 }
118 
119 void
120 ZfsDaemon::RequestSystemRescan()
121 {
122 	s_systemRescanRequested = true;
123 	ZfsDaemon::WakeEventLoop();
124 }
125 
126 void
127 ZfsDaemon::Run()
128 {
129 	ZfsDaemon daemon;
130 
131 	while (s_terminateEventLoop == false) {
132 
133 		try {
134 			daemon.DisconnectFromDevd();
135 
136 			if (daemon.ConnectToDevd() == false) {
137 				sleep(30);
138 				continue;
139 			}
140 
141 			daemon.DetectMissedEvents();
142 
143 			daemon.EventLoop();
144 
145 		} catch (const DevdCtl::Exception &exp) {
146 			exp.Log();
147 		}
148 	}
149 
150 	daemon.DisconnectFromDevd();
151 }
152 
153 //- ZfsDaemon Private Methods --------------------------------------------------
154 ZfsDaemon::ZfsDaemon()
155  : Consumer(/*defBuilder*/NULL, s_registryEntries,
156 	    NUM_ELEMENTS(s_registryEntries))
157 {
158 	if (s_theZfsDaemon != NULL)
159 		errx(1, "Multiple ZfsDaemon instances created. Exiting");
160 
161 	s_theZfsDaemon = this;
162 
163 	if (pipe(s_signalPipeFD) != 0)
164 		errx(1, "Unable to allocate signal pipe. Exiting");
165 
166 	if (fcntl(s_signalPipeFD[0], F_SETFL, O_NONBLOCK) == -1)
167 		errx(1, "Unable to set pipe as non-blocking. Exiting");
168 
169 	if (fcntl(s_signalPipeFD[1], F_SETFL, O_NONBLOCK) == -1)
170 		errx(1, "Unable to set pipe as non-blocking. Exiting");
171 
172 	signal(SIGHUP,  ZfsDaemon::RescanSignalHandler);
173 	signal(SIGINFO, ZfsDaemon::InfoSignalHandler);
174 	signal(SIGINT,  ZfsDaemon::QuitSignalHandler);
175 	signal(SIGTERM, ZfsDaemon::QuitSignalHandler);
176 	signal(SIGUSR1, ZfsDaemon::RescanSignalHandler);
177 
178 	g_zfsHandle = libzfs_init();
179 	if (g_zfsHandle == NULL)
180 		errx(1, "Unable to initialize ZFS library. Exiting");
181 
182 	Callout::Init();
183 	InitializeSyslog();
184 	OpenPIDFile();
185 
186 	if (g_debug == 0)
187 		daemon(0, 0);
188 
189 	UpdatePIDFile();
190 }
191 
192 ZfsDaemon::~ZfsDaemon()
193 {
194 	PurgeCaseFiles();
195 	ClosePIDFile();
196 }
197 
198 void
199 ZfsDaemon::PurgeCaseFiles()
200 {
201 	CaseFile::PurgeAll();
202 }
203 
204 bool
205 ZfsDaemon::VdevAddCaseFile(Vdev &vdev, void *cbArg)
206 {
207 	if (vdev.State() != VDEV_STATE_HEALTHY)
208 		CaseFile::Create(vdev);
209 
210 	return (/*break early*/false);
211 }
212 
213 void
214 ZfsDaemon::BuildCaseFiles()
215 {
216 	ZpoolList zpl;
217 	ZpoolList::iterator pool;
218 
219 	/* Add CaseFiles for vdevs with issues. */
220 	for (pool = zpl.begin(); pool != zpl.end(); pool++)
221 		VdevIterator(*pool).Each(VdevAddCaseFile, NULL);
222 
223 	/* De-serialize any saved cases. */
224 	CaseFile::DeSerialize();
225 
226 	/* Simulate config_sync events to force CaseFile reevaluation */
227 	for (pool = zpl.begin(); pool != zpl.end(); pool++) {
228 		char evString[160];
229 		Event *event;
230 		nvlist_t *config;
231 		uint64_t poolGUID;
232 		const char *poolname;
233 
234 		poolname = zpool_get_name(*pool);
235 		config = zpool_get_config(*pool, NULL);
236 		if (config == NULL) {
237 			syslog(LOG_ERR, "ZFSDaemon::BuildCaseFiles: Could not "
238 			    "find pool config for pool %s", poolname);
239 			continue;
240 		}
241 		if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
242 				     &poolGUID) != 0) {
243 			syslog(LOG_ERR, "ZFSDaemon::BuildCaseFiles: Could not "
244 			    "find pool guid for pool %s", poolname);
245 			continue;
246 		}
247 
248 
249 		snprintf(evString, 160, "!system=ZFS subsystem=ZFS "
250 		    "type=misc.fs.zfs.config_sync sub_type=synthesized "
251 		    "pool_name=%s pool_guid=%" PRIu64 "\n", poolname, poolGUID);
252 		event = Event::CreateEvent(GetFactory(), string(evString));
253 		if (event != NULL) {
254 			event->Process();
255 			delete event;
256 		}
257 	}
258 }
259 
260 void
261 ZfsDaemon::RescanSystem()
262 {
263         struct gmesh	  mesh;
264         struct gclass	 *mp;
265         struct ggeom	 *gp;
266         struct gprovider *pp;
267 	int		  result;
268 
269         /*
270 	 * The devdctl system doesn't replay events for new consumers
271 	 * of the interface.  Emit manufactured DEVFS arrival events
272 	 * for any devices that already before we started or during
273 	 * periods where we've lost our connection to devd.
274          */
275 	result = geom_gettree(&mesh);
276 	if (result != 0) {
277 		syslog(LOG_ERR, "ZfsDaemon::RescanSystem: "
278 		       "geom_gettree faild with error %d\n", result);
279 		return;
280 	}
281 
282 	const string evStart("!system=DEVFS subsystem=CDEV type=CREATE "
283 			     "sub_type=synthesized cdev=");
284         LIST_FOREACH(mp, &mesh.lg_class, lg_class) {
285                 LIST_FOREACH(gp, &mp->lg_geom, lg_geom) {
286                         LIST_FOREACH(pp, &gp->lg_provider, lg_provider) {
287 				Event *event;
288 
289 				string evString(evStart + pp->lg_name + "\n");
290 				event = Event::CreateEvent(GetFactory(),
291 							   evString);
292 				if (event != NULL) {
293 					if (event->Process())
294 						SaveEvent(*event);
295 					delete event;
296 				}
297                         }
298                 }
299 	}
300 	geom_deletetree(&mesh);
301 }
302 
303 void
304 ZfsDaemon::DetectMissedEvents()
305 {
306 	do {
307 		PurgeCaseFiles();
308 
309 		/*
310 		 * Discard any events waiting for us.  We don't know
311 		 * if they still apply to the current state of the
312 		 * system.
313 		 */
314 		FlushEvents();
315 
316 		BuildCaseFiles();
317 
318 		/*
319 		 * If the system state has changed during our
320 		 * interrogation, start over.
321 		 */
322 	} while (s_terminateEventLoop == false && EventsPending());
323 
324 	RescanSystem();
325 }
326 
327 void
328 ZfsDaemon::EventLoop()
329 {
330 	while (s_terminateEventLoop == false) {
331 		struct pollfd fds[2];
332 		int	      result;
333 
334 		if (s_logCaseFiles == true) {
335 			EventList::iterator event(m_unconsumedEvents.begin());
336 			s_logCaseFiles = false;
337 			CaseFile::LogAll();
338 			while (event != m_unconsumedEvents.end())
339 				(*event++)->Log(LOG_INFO);
340 		}
341 
342 		Callout::ExpireCallouts();
343 
344 		/* Wait for data. */
345 		fds[0].fd      = m_devdSockFD;
346 		fds[0].events  = POLLIN;
347 		fds[0].revents = 0;
348 		fds[1].fd      = s_signalPipeFD[0];
349 		fds[1].events  = POLLIN;
350 		fds[1].revents = 0;
351 		result = poll(fds, NUM_ELEMENTS(fds), /*timeout*/INFTIM);
352 		if (result == -1) {
353 			if (errno == EINTR)
354 				continue;
355 			else
356 				err(1, "Polling for devd events failed");
357 		} else if (result == 0) {
358 			errx(1, "Unexpected result of 0 from poll. Exiting");
359 		}
360 
361 		if ((fds[0].revents & POLLIN) != 0)
362 			ProcessEvents();
363 
364 		if ((fds[1].revents & POLLIN) != 0) {
365 			static char discardBuf[128];
366 
367 			/*
368 			 * This pipe exists just to close the signal
369 			 * race.  Its contents are of no interest to
370 			 * us, but we must ensure that future signals
371 			 * have space in the pipe to write.
372 			 */
373 			while (read(s_signalPipeFD[0], discardBuf,
374 				    sizeof(discardBuf)) > 0)
375 				;
376 		}
377 
378 		if (s_systemRescanRequested == true) {
379 			s_systemRescanRequested = false;
380 			syslog(LOG_INFO, "System Rescan request processed.");
381 			RescanSystem();
382 		}
383 
384 		if ((fds[0].revents & POLLERR) != 0) {
385 			syslog(LOG_INFO, "POLLERROR detected on devd socket.");
386 			break;
387 		}
388 
389 		if ((fds[0].revents & POLLHUP) != 0) {
390 			syslog(LOG_INFO, "POLLHUP detected on devd socket.");
391 			break;
392 		}
393 	}
394 }
395 //- ZfsDaemon staic Private Methods --------------------------------------------
396 void
397 ZfsDaemon::InfoSignalHandler(int)
398 {
399 	s_logCaseFiles = true;
400 	ZfsDaemon::WakeEventLoop();
401 }
402 
403 void
404 ZfsDaemon::RescanSignalHandler(int)
405 {
406 	RequestSystemRescan();
407 }
408 
409 void
410 ZfsDaemon::QuitSignalHandler(int)
411 {
412 	s_terminateEventLoop = true;
413 	ZfsDaemon::WakeEventLoop();
414 }
415 
416 void
417 ZfsDaemon::OpenPIDFile()
418 {
419 	pid_t otherPID;
420 
421 	s_pidFH = pidfile_open(s_pidFilePath, 0600, &otherPID);
422 	if (s_pidFH == NULL) {
423 		if (errno == EEXIST)
424 			errx(1, "already running as PID %d. Exiting", otherPID);
425 		warn("cannot open PID file");
426 	}
427 }
428 
429 void
430 ZfsDaemon::UpdatePIDFile()
431 {
432 	if (s_pidFH != NULL)
433 		pidfile_write(s_pidFH);
434 }
435 
436 void
437 ZfsDaemon::ClosePIDFile()
438 {
439 	if (s_pidFH != NULL)
440 		pidfile_close(s_pidFH);
441 }
442 
443 void
444 ZfsDaemon::InitializeSyslog()
445 {
446 	openlog("zfsd", LOG_NDELAY, LOG_DAEMON);
447 }
448 
449